COSMO-Agent: LLM Framework for Closed-Loop CAD-CAE Optimization
Researchers have introduced COSMO-Agent (Closed-loop Optimization, Simulation, and Modeling Orchestration), a framework that enhances reinforcement learning with tools to train large language models (LLMs) in addressing the semantic gap between CAD and CAE during iterative optimization in industrial design and simulation. This framework treats the processes of CAD generation, CAE solving, result interpretation, and geometry modification as an interactive reinforcement learning environment. Here, an LLM learns to coordinate external tools and adjust parametric geometries until all constraints are met. A multi-constraint reward system promotes feasibility, robustness of the toolchain, and validity of structured outputs. The study features a dataset aligned with industry needs, encompassing 25 component categories with functional CAD-CAE models. The paper can be found on arXiv under ID 2605.20190.
Key facts
- COSMO-Agent is a tool-augmented reinforcement learning framework for LLMs
- It addresses the CAD-CAE semantic gap in industrial design-simulation optimization
- The framework casts CAD generation, CAE solving, result parsing, and geometry revision as an interactive RL environment
- A multi-constraint reward encourages feasibility, toolchain robustness, and structured output validity
- An industry-aligned dataset covers 25 component categories with executable CAD-CAE models
- The paper is available on arXiv with ID 2605.20190
Entities
Institutions
- arXiv